Image Reconstruction through Generative Adversarial Networks

For many years, inpainting has been a focus in the field of computer vision. To be able to reconstruct damaged portions of an image has numerous applications, from realistically repairing photographs to smoothly removing unwanted areas in digital image editing. Recently, the Generative Adversarial Network (GAN) has shown to perform well when tasked with recreating natural images. In this paper, we aim to evaluate the GAN in comparison to a direct convolutional network, for the task of recreating missing areas of various images.